The work is based on data about the prevalence of congenital anomalies among newborns in Lithuania. The log-linear model is used to assess dependence structure of a subset of categorical variables. It is shown that fitting the log-linear model with just three categorical variables can be a rather complicated task due to large number of unknown parameters and cells in the contingency table. The classical chi-squre test and the bootstrap technique are compared for testing goodness-of-fit. The results demonstrate that the number of cells of even nonsparse contingency tables has significant impact on the tail distribution of the likelihood ratio statistics.
This work is licensed under a Creative Commons Attribution 4.0 International License.